Predicting Neck Fluid Accumulation While Supine · leg fluid volume (LFV) between 100 and 300 ml [14–16]. When lying down, the reverse occurs; hydrostatic pressure in the capillaries

Journal of Healthcare Engineering · Vol. 6 · No. 4 · 2015 Page 673–690 673

Predicting Neck Fluid Accumulation While SupineDaniel Vena, MHSc1,2; Babak Taati, PhD1,3 and Azadeh Yadollahi, PhD1,2*

1University Health Network – Toronto Rehabilitation Institute, Toronto, Canada2Institute of Biomaterials and Biomedical Engineering,

University of Toronto, Toronto, Canada3Department of Computer Science, University of Toronto, Toronto, Canada

Submitted May 2015. Accepted for publication July 2015.

ABSTRACTWhen lying supine, fluid shifts rostrally from the legs and accumulates in the neck, which is arisk factor for obstructive sleep apnea. The objective of this study was to model neck fluidaccumulation using one-time baseline measurements of body fluid, demographics, andanthropometrics. Using bioelectrical impedance, leg and neck fluid volumes (LFV and NFV)were measured continuously and simultaneously. Thirty non-obese adults (13 men) stood quietlyfor 5 minutes, and then lay supine for 90 minutes while fluid volumes were measured. Neckcircumference (NC) was measured before and after the supine period. Results demonstrated that,compared to women, men experienced a greater increase in NC after lying supine. Furthermore,baseline LFV at the onset of lying supine was significantly correlated with ΔLFV (r = 0.44, p = 0.014) and ΔNC (r = 0.51, p = 0.008) after 90 minutes supine. The findings identify that sexand baseline LFV predict both the fluid leaving the legs and increase in NC during recumbency.

Keywords: Rostral fluid shift, leg fluid volume, neck fluid volume, supine position, upper airway

1. INTRODUCTIONOvernight neck fluid accumulation is a risk factor for upper airway pathologies andrespiratory disorders [1]–[4]. The accumulation of neck fluid could cause distension ofthe neck veins and/or edema formation in the pharyngeal soft tissue. These changesincrease pressure on the upper airway and could cause the narrowing of the upperairway [3, 5, 6], increase upper airway resistance [1,7] and collapsibility [4,8], andincrease the severity of obstructive sleep apnea (OSA) [9]. Understanding the factorsthat increase the likelihood of fluid accumulation in the neck can help identify patientsat risk for fluid related respiratory disorders and also identify avenues for new therapiesthat could target these factors. The redistribution of fluid from the legs into the upperbody while recumbent is an important mechanism through which fluid may accumulatein the neck. It has been demonstrated that the volume of fluid leaving the legs whilerecumbent is correlated with increased neck circumference (NC) signaling theaccumulation of fluid in the neck [2, 10, 11].

*Corresponding author: Azadeh Yadollahi, University Hospital Network – Toronto Rehabilitation Institute,Room 12-106, 550 University Ave., Toronto, ON, M5G 2A2, Canada. Telephone: 416-597-3422 x7936. Fax:416-597-8959. Email: [email protected]. Other authors: [email protected], [email protected].

674 Predicting Neck Fluid Accumulation While Supine

The effects of gravity and Starling’s forces cause fluid redistribution during thetransition from a standing to a recumbent position. Based on the Starling model, thebalance of hydrostatic and oncotic forces controls the fluid movement between the capillaries and the interstitial spaces [12]. When hydrostatic pressure is higher in thecapillary spaces compared to the interstitial spaces, e.g., in sitting or standing postures, theresulting pressure gradient promotes the movement of blood plasma out of the capillariesand into the interstitial spaces [13]. Therefore, these postures result in an increased overallleg fluid volume (LFV) between 100 and 300 ml [14–16]. When lying down, the reverseoccurs; hydrostatic pressure in the capillaries of the legs is reduced, allowing interstitialfluid to be reabsorbed back into the venous system [17–20], which moves rostrally throughthe vascular system to the upper body due to gravity. The result is reduced fluid volume inthe legs and associated increases in fluid volumes of the thorax and head [18, 19].

Identifying patients that are susceptible to increased neck fluid accumulation afterprolonged recumbency can be a useful predictor of individuals at risk of upper airwaypathologies due to fluid shift. However, measuring the rostral shift of fluid from the legsto the neck during prolonged recumbency is inconvenient and time consuming. It eitherrequires data collection for hours or overnight [2], or the use of anti-shock trousers toapply lower body positive pressure to quicken LFV shift and simulate prolonged rostralfluid shift [1], which can be cumbersome. Therefore, it is important to investigatesimple measurements that can be performed quickly and conveniently (e.g., in clinic),which in turn could predict the amount of fluid that would accumulate in the neck afterprolonged recumbency. Throughout the paper, we refer to these instantaneous, one-timemeasurements taken before prolonged recumbency as baseline measures.

Previous studies have shown that in heart failure patients, an increased baseline legedema score is significantly correlated with the overnight change in LFV and theincrease in NC [11,21]. However, leg edema score is determined subjectively; it involvesthe application of pressure to a small area on the lower leg to create an indentation in thetissue. A subjective score from 1+ (less severe) to 4+ (more severe) is given based on thedepth of indentation and how long the indentation remains [22]. In addition, these studiesonly focused on heart failure patients experiencing fluid overload.

The current study explored various baseline metrics measured objectively toimprove prediction of neck fluid accumulation in the general population. These metricsincluded: baseline LFV, NC, neck fluid volume (NFV), and upper-airway cross-sectional area (UAXSA). Furthermore, in past studies, neck fluid accumulation hadonly been measured using the change in NC as a surrogate measure [2, 10, 11]. In thisstudy, the bioelectrical impedance method was used to simultaneously measure bothLFV and NFV [23]. The purpose of the present study was to develop a model based onbaseline measurements of body fluid, as well as demographics and anthropometrics ofthe participants to predict fluid accumulation in the neck.

2. MATERIALS AND METHODS2.1. ParticipantsThe protocol was approved by the Research Ethics Board of Toronto RehabilitationInstitute, and all participants provided written informed consent prior to participating inthe study. Criteria for inclusion were men and women between 18 and 65 years of age

Journal of Healthcare Engineering · Vol. 6 · No. 4 · 2015 675

with a body mass index (BMI) <30 kg/m2, and a blood pressure ≤140/90 mmHg.Women were included only if they were premenopausal and did not have theirmenstrual period at the time of experiments. The exclusion criteria were a history ofhysterectomy, having metal implants, history of cardiovascular, renal, or respiratorydiseases, use of prescribed medications for those diseases, or taking any over thecounter medication that might influence fluid retention, such as diuretics or non-steroidal anti-inflammatory agents. Participants were instructed to avoid the intake ofalcohol or caffeine on the day of the study. Participants were recruited from thecommunity by advertisement.

2.2. Neck Circumference, Upper-Airway Cross-Sectional Area, and Airway LengthMeasurementsAt the beginning of each session, a tape measure was used to measure NC just abovethe cricothyroid cartilage. A line was drawn at this level to ensure the NC measurementat the end of the experiment was made at the same level, as described in past studies [4,21]. Upper-airway length (LUA, distance from velum to glottis) and the averageUAXSA from velum to glottis were measured using acoustic pharyngometry [24].

2.3. Leg and Neck Fluid Volumes MeasurementsFluid was measured in the leg and neck using bioelectrical impedance, a non-invasivetechnique used to estimate the fluid volume of tissues. The bioelectrical impedancemethod of fluid measurement is well validated and highly reproducible with anaccuracy of 0.5% compared to reference measures of total body water, repeatabilitywithin 0.3%, and test-retest correlation >95% [25,26].

The method is based on Ohm’s law (V = IR), where the resistance of a tissue toelectrical current is inversely related to its fluid content and directly related to its length:R = rL2/v, where r is the resistivity of the fluid, L is the segment’s length, v is the fluidvolume, and R is resistance [27–29]. The resulting equation (v = rL2/R) has been widelyused to estimate total body water, where L is replaced with the subject’s height [25,26].Past studies have also used bioelectrical impedance to measure the fluid volume ofindividual body segments. In these studies, segments are assumed to be cylindrical inshape, with L representing the length of the segment [30–33]. In a previous study, wedeveloped a system for continuous measurement of fluid volumes in various bodysegments [34]. We used a modified version of this equation to reflect the tapered shapeof the body segments (leg, abdomen, chest, and neck). Therefore, fluid volume wasestimated as [35], [36]:

(1)

where C1 and C2 are the top and bottom circumferences of the segment, respectively, Lis the segment’s length, R is the segment’s resistance, and ρ is blood resistivity, whichis estimated as 47 Ωcm [29].

ρ ( )( )=

⎛

⎝⎜⎜⎜

⎞

⎠⎟⎟⎟ + +v

3 4π

L

C C R L C C C C

2/3

1/31 2

2/3

12

22

1 2

For both the leg and the neck, two surface electrodes were used to inject a lowamplitude (400 μa), high frequency current (25 kHz and 50 kHz for the leg and neck,respectively) and two surface electrodes were used to measure the voltage (Figure 1)to estimate bioelectrical impedance of the segment (R in eqn. 1) [34]. In Figure 1,electrodes to inject current are denoted as I– and I+ and electrodes to measure voltage


Figure 1. Positions of the surface electrodes for recording the bioelectricalimpedance in the leg and neck. I+ and I– denote surface electrodes usedto inject a low amplitude (400 μa), high frequency (25 kHz and 50 kHzfor the leg and neck, respectively) current. V+ and V– denote surfaceelectrodes used to measure the voltages across the leg and neck segments.LFV denotes leg fluid volume and NFV denotes neck fluid volume.

I+

I+

V+

V+

V-

V-

NFV

I-

I-

LFV

are denoted as V– and V+. Voltage measuring electrodes were placed at the ankle andupper thigh of the right leg for the LFV measurement. To measure NFV, voltage-measuring electrodes were placed on the right side of the neck below the right ear andat the base of the neck (Figure 1). The current-injecting electrodes were placed oneinch away from the voltage-measuring electrodes. To isolate the signal from differentbody segments for bioelectrical impedance measurements, a different injectingfrequency was used for the leg and neck. The electrodes were secured to the skinusing adhesive tape. At the beginning of the study, length and circumference of eachsegment were measured with a measuring tape at the level of the voltage-measuringelectrodes. For NFV, the NC measurement was used for both circumferences C1 andC2 (eqn. 1).

2.4. ProtocolThis study is part of a previous study [34] conducted at the Toronto RehabilitationInstitute. While seated, blood pressure was measured to ensure subjects werenormotensive. Next, surface electrodes were applied to the participant, as shown inFigure 1, to measure the bioelectrical impedance. All experiments were performed inthe same room with the temperature maintained between 22°C and 24°C. Next,participants were asked to stand motionless for 5 minutes. For the following 90 minutes,the subjects then lay awake in the supine position on a bed without a pillow.Bioelectrical impedance of the leg and neck were recorded continuously andsimultaneously during both standing and supine positions. Subjects were instructed toremain motionless during the recordings.

2.5. Baseline and Outcome VariablesBaseline variables were selected as quantities that could be easily measured and representfluid content in the body and/or have been shown to have a strong correlation withincreases in NC while supine. The variables included were LFV after standing for 5minutes (LFVst5), NFV after standing for 5 minutes (NFVst5), LFV at baseline supine(LFVsp1), NFV at baseline supine (NFVsp1), immediate change in LFV upon transitioningfrom standing to supine position (ΔLFVP = LFVsp1 – LFVst5), immediate change in NFVupon transitioning from standing to supine position (ΔNFVP = NFVsp1 – NFVst5), NCmeasured at baseline supine (NCsp1), upper-airway cross-sectional area at baseline supine(UAXSAsp1), and length of the upper-airway (between the velum and the glottis)measured at baseline supine (LUA). Also computed was ΔLFV90 (ΔLFV90 = LFVsp90 –LFVsp1), where LFVsp90 is the LFV after 90 minutes supine. Since previous studies haveshown strong correlations between the change in LFV and change in NC [1]–[4], ΔLFV90

was included to correlate independently with the outcome variables and to compare ourresults with those of the previous studies; however, it was not included in the multipleregression models, as discussed later. Lastly, demographic information including age, sex,height, weight, and BMI were also included in the regression model. Sex was defined asa binary variable with 0 and 1 representing men and women, respectively.

Outcome variables were selected as ΔNC90 = NCsp90 – NCsp1, ΔNFV90 = NFVsp90 –NFVsp1, and ΔUAXSA90 = UAXSAsp90 – UAXSAsp1, where NCsp90, NFVsp90, andUAXSAsp90 were the NC, NFV and UAXSA after lying supine for 90 minutes, respectively.


2.6. Data AnalysisThe analysis was performed in three steps. In the first step, individual correlations wereanalyzed between each of the baseline variables and the outcome variables usingPearson correlations with normally distributed data, and Spearman’s rank correlationwith non-normally distributed data. Normality of the data was determined using theKolmogorov-Smirnov test. A correlation was considered significant with a two-sided p-value < 0.05.

In the second step of the analysis, a multiple linear regression model was developedto determine the most significant factors that contributed to the changes in outcomevariables. Baseline variables were selected for the model using a forward stepwiseselection. Using this approach, the model starts with no variables from the variable set,adding a variable if their associated p-value is below 0.05, and removing variables iftheir p-value is above 0.10. The variable set included the entire baseline variable setalready described, excluding ΔLFVP and ΔNFVP, since they are linear combinations ofthe baseline leg and neck fluid volume measurements in the supine and standingposition. We also excluded weight and height in the same variable set as BMI, giventhat BMI is a function of weight and height.

The final step of the analysis was to investigate whether the subjects can beclassified into two groups of “high risk” and “low risk” regarding the effects of fluidaccumulation in the neck. To do this, input variables were converted into standardizedz-scores and a method based on the combination of principal component analysis(PCA) and unsupervised clustering was implemented. PCA is a statistical method thatconverts a set of correlated variables into a new set of uncorrelated (orthogonal)variables called principal components [37]. The principal components are linearcombinations of the original variables weighted by their contribution to explaining thevariance in a particular orthogonal dimension. PCA was used to reduce thedimensionality of an input variable set and to identify a subset of input variables thataccount for most of the variability in the data. The number of components to retain wasselected based on the point on inflexion on the Scree plot (Figure 2a) which plots thecomponent number versus the Eigenvalue as the dependent variable. The Eigenvalue isa scalar indicator of the substantive importance of the associated component with largerEigenvalues representing more important components. Components were retained ifthey were to the left of the point of inflexion, but not including the inflexion point itself.We applied a similar approach for selecting input variables by plotting the factorweightings of each input variable in descending order (Figure 2b). A point of inflexionwas identified and variables to the left of the point of inflexion were retained as avariable subset for further clustering analysis.

An unsupervised K-means clustering was then applied to the selected subset ofvariables to cluster the subjects into two groups (K = 2). This clustering methodseparates the data into K groups by establishing a centroid for each group. The positionfor the centroid is initialized and data points are assigned to the closest centroid (theassignment step). The centroid is then repositioned to minimize the within-cluster sumof squares (the update step). The assignment step and the update step are repeated untilconvergence; when the assignments no longer change [38]. Outcome variables were



4.5

0.45

0.4

0.35

0.3

Wei

ghtin

g

0.26

0.2

0.16

0.1

0.05

LFV st

5LF

V sp1

NFV

sp1

NFV

st5

Nc sp

1

L UA

∆LFv

p

∆NFV

p p

UAX

SAsp

1

Sex

BM1

Age

(a)

(b)

4

3.5

3

2.5E

igen

valu

e

2

1.5

1

0.5

00 2 4 6

Component

Variable

8 10 12

Figure 2. (a) Scree plot of the Eigenvalues of each component used to identify theinflexion point used as a threshold for the number of components toretain. The point of inflexion is located at Component 2, which meansthat only the first (Component) is retained. (b) A modified scree plot ofthe variables and the weightings of the variables on the first component,similarly used to identify the point of inflexion as a threshold for numberof variables to retain. The inflexion point is located at the BMI variable,meaning all variables to the left of BMI are retained.

compared between the two clusters using the student t-test for normally distributed dataand the Kolmogorov-Smirnov test for non-normally distributed data; a two-sided p-value of less than 0.05 was considered significant. Statistical analysis was performedusing SAS 9.3 (SAS Institute Inc., Cary, NC), and PCA and clustering were performedin Matlab (MathWorks, Natick, MA).

3. RESULTSFifty-two candidates consented to participate in the study. One participant declined tocontinue after being instrumented. Eleven participants had movement artifacts in thebioelectrical impedance signals and their data were excluded. Ten participants did nothave standing data collected and were excluded from this analysis. After all exclusions,a total of 30 participants (13 men and 17 women) were included in the analysis.Baseline fluid measures, as well as anthropometric data and variables for men, women,and all subjects are shown in Table 1. After lying supine for 90 minutes, NFV and NCin all subjects increased significantly (p < 0.0001), while the upper-airway cross-sectional area reduced significantly (p = 0.036). Compared to men, women had smallerweight, height, baseline NC and UA length, but had similar age and BMI. In addition,women had lower baseline NFV and LFV at standing and supine. However, bothexperienced a similar change in LFV over the 90-minute supine period. In terms of


Table 1. Baseline and outcome variable data in all subjects, and associated p-values for statistical differences between men and women

P-value between

Variable All Men (n=13) Women (n=17) sexes

Age, years 39.2±11.7 37.8±13.2 39.2±11.7 0.771BMI, kg/m2 23.3±2.9 25.0±4.3 22.8±2.6 0.771Weight, kg 69.0±11.8 81.7±15.7 63.0±9.5 <0.001Height, cm 172.0±10.7 180.7±9.3 165.9±7.5 <0.001NCsp1, cm 36.6±3.8 41.0±3.4 34.0±2.0 <0.001UAXSAsp1, cm2 2.9±1.4 3.0±1.5 2.7±1.1 0.242NFVst5, ml 223.9±60.2 253.8±72.5 200.7±26.6 0.009LFVst5, ml 2195.5±288.9 2532.2±482.6 2057.5±256.4 0.006ΔNFVP, ml 12.1±8.9 11.0±12.5 11.9±7.6 0.685ΔLFVP, ml 66.7±28.7 59.6±51.5 68.0±21.0 0.415NFVsp1, ml 236.0±60.5 264.8±71.9 212.6±28.1 0.006LFVsp1, ml 2128.8±278.7 2472.7±499.2 1989.8±246.4 0.002ΔLFV90, ml 143.7±37.3 160.9±56.6 138.9±32.4 0.182LUA, cm 8.8±1.5 10.1±1.6 8.1±0.9 0.004ΔNC90, cm 0.5±0.4 0.7±0.3 0.4±0.3 0.013ΔNFV90, ml 15.7±4.4 16.8±5.2 14.3±3.1 0.082ΔUAXSA90, cm2 −0.3±0.7 −0.4±0.7 −0.2±0.7 0.136

outcome variables, women experienced a smaller change in NC after 90 minutes supinecompared to men (p = 0.013). There was also a tendency for an increased change inNFV after 90 minutes supine in men compared to women (p = 0.087). However, thedifferences between women and men in the change in UAXSA after 90 minutes supinewas not statistically significant.

Individual correlations between baseline variables and outcome variables aresummarized in Table 2. Three baseline variables were significantly correlated with thechange in NC after 90 minutes supine: sex (r = 0.56, p = 0.001), and baseline LFV atstanding (r = 0.48, p = 0.014) and supine (r = 0.51, p = 0.008).

These correlations demonstrate that male sex and increased baseline LFVcorresponded to an increase in NC over the 90-minute supine period. Change in NFVafter lying supine for 90 minutes was directly and significantly correlated with baselineNFV at standing (r = 0.41, p = 0.026) and baseline UAXSA (r = 0.38, p = 0.042). It washardly significant with NFV at supine (r = 0.36, p = 0.052). These same variables werealso significantly correlated to the narrowing in the UAXSA after lying supine (NFVst5:r = −0.44, p = 0.018, NFVsp1: r = −0.46, p = 0.011, UAXSAsp1: r = −0.48, p = 0.0078).Lastly, DLFV90 was significantly correlated with baseline LFV at standing (r = 0.39, p = 0.032) and supine (r = 0.44, p = 0.014). In terms of the correlation across outcomevariables, the change in NC was significantly correlated with the change in NFV after


Table 2. Correlation coefficients for both the individual baseline variables andthe multiple linear models developed from stepwise regression

Correlation

Variables ΔNC90 ΔNFV90 ΔUAXSA90

Sex −0.561* −0.280 0.191BMI 0.106 0.144 0.068Weight 0.230 0.124 −0.112Height 0.241 0.288 −0.206Age 0.263 −0.229 0.066NCsp1 0.314 0.089 −0183UAXSAsp1 −0.022 0.380* −0.484*

NFVst5 −0.046 0.406* −0.437*

LFVst5 0.483* 0.281 −0.002ΔNFVP −0.346 −0.167 0.083ΔLFVP −0.091 0.002 −0.176NFVsp1 −0.106 0.357Ü −0.463*

LFVsp1 0.510* 0.292 0.014ΔLFV90 0.348 0.306 0.069LUA 0.160 0.121 −0.281Model 0.687* 0.188* 0.234*

Statistical significance with p < 0.05 is denoted with superscript *, and borderline significance with p = 0.05is denoted with a superscript Ü.

90 minutes supine (r = 0.364, p = 0.047). It was not correlated with the change UA-XSAafter 90 minutes supine (p = 0.82). In addition, the correlation between change in NFVand the change in UA-XSA after 90 minutes supine was not statistically significant (r =−0.323, p=0.087).

The linear regression models developed using stepwise feature selections for eachoutcome variable are summarized in Table 3. Models for ΔNFV90 and ΔUAXSA90 onlyincluded one variable (NFVsp1 and UAXSAsp1, respectively). The linear regressionmodel of ΔNC90 showed that sex, NFVsp1, LFVst5, and NCsp1 contributed to theincreased NC after lying supine for 90 minutes. Models developed with weight andheight replacing BMI in the baseline variable set did not change the models and werenot reported. The correlation between the linear regression models and the outcomevariables are listed in Table 2. Since the models developed for ΔNFV90 andΔUAXSA90

included only one variable, the correlations were unchanged compared to those fromsingle-variable models. On the other hand, the correlation of the proposed model withΔNC90 was higher than the correlations of individual variables with ΔNC90.

Only the first principle component from the PCA was retained, since the point ofinflexion was positioned at the second component (Figure 2a). The first principlecomponent explained 37.1% of the variance in the data set. The point of inflexion in themodified Scree plot of the first principal component (Figure 2b) was positioned aroundBMI. Thus, sex, baseline NC, and baseline NFV/LFV at both standing and supinepositions were retained for further analysis. Given the co-linearity of the standing andsupine measures of LFV and NFV, only standing LFV and NFV were included in thevariable subset. Clustering based on these variables yielded two distinct groups. Thedominant factor that separated the clusters was sex, so cluster 1 contained only men,while cluster 2 contained only women. To explore the data further, sex was removedfrom the variable subset and K-means clustering was repeated.

These results are illustrated in Figure 3. As shown in Figure 3, clusters 1 and 2 arestill separated based mostly on sex, with cluster 2 consisting of only women, and cluster1 consisting of majority men and two women. The clusters show a good separationacross all of the input variables NCsp1, NFVst5, and LFVst5, which were significantlydifferent between clusters (see Table 4). Differences in the value of the outcomevariables of ΔNC90, ΔNFV90 , and ΔUAXSA90 between clusters 1 and 2 are shown in


Table 3. Results of the stepwise regression for each outcome variable

P-value in theDependent Variable Step Variable entered Partial R2 Model R2 final model

ΔNC90 1 Sex 0.293 0.293 0.0022 NFVsp1 0.204 0.497 <0.0013 LFVst5 0.117 0.614 <0.0014 NCsp1 0.073 0.687 <0.001

ΔNFV90 1 NFVsp1 0.188 0.188 0.020ΔUAXSA90 1 UAXSAsp1 0.234 0.234 0.008


Centroid

5(a)

(b)2

1.5

1

0.5

0

−0.5

−1

−1.5

−2−2 −1 0 1 2

Baseline NFV (z-score)

Bas

elin

e LF

V (

z-sc

ore)

3 4 5

4

3

2

1

0

−1

−2−1.5 −1 −0.5 0

Baseline NC (z-score)

Bas

elin

e N

FV

(z-

scor

e)

0.5 1 1.5 2

Cluster 1Cluster 2WomenMen

Figure 3. Plots of clusters and centroid locations on the two-dimensional plot of (a)baseline neck fluid volume (NFV) and baseline neck circumference(NC), and (b) baseline leg fluid volume (LFV) and baseline neck fluidvolume (NFV). Clustering based on the variable subset that includedbaselines NC (NCsp1), baseline standing LFV (LFVst5), and NFV(NFVst5) yielded two distinct clusters, with cluster 2 consisting of onlywomen, and cluster 1 consisting of majority men. The input variablesNCsp1, NFVst5, and LFVst5, were significantly different between clusters,but not the outcome variables including the change in NC, NFV andupper-airway cross-sectional area after 90 minutes supine (ΔNC90,ΔNFV90 and ΔUAXSA90, respectively).

Table 4. There was a non-significant tendency for the participants of cluster 2 to have ahigher increase in ΔNC90, ΔNFV90 and a greater reduction in ΔUAXSA90 compared tocluster 1.

4. DISCUSSIONThe most important finding of our study was that demographic information andbaseline measures of fluid in the body could be used to predict the amount of fluidleaving the legs and the increase in neck circumference after lying down for 90 minutes.For the first time, we showed that baseline standing and supine LFV have a positive andsignificant correlation with changes in LFV and NC after lying supine for 90 minutes.Previous studies have shown similar findings that subjective clinical measures of legedema score before sleep is correlated with the overnight change in LFV and increasedseverity of sleep apnea [11,21]. It has also been shown that the change in LFV duringsleep is the strongest predictor of the change in NC, as well as the consequent increasein the severity of sleep apnea [2,21]. These studies show strong evidence of theimportance of the overnight change in LFV as a predictor of fluid accumulation in theneck and sleep apnea severity. However, they are based on measurements of LFVacquired overnight or during a prolonged period of recumbency. Our results are uniqueto suggest that a simpler objective measure of LFV taken at baseline could be used asa strong predictor of the amount of fluid that will leave the legs, causing an increase inNC during a prolonged period of recumbency.

Another demographic that predicted the change in NC was sex. Although change inLFV was similar between the sexes, men experienced a greater increase in the NCduring a supine period. The effect of sex on rostral fluid shift has been previouslyshown in overnight studies in patients with heart failure. These studies showed that theovernight change in LFV in men was significantly correlated to the increase in NC[11]. For women with the same amount of fluid leaving the legs, an overnight changein LFV did not correlate with the overnight change in NC [11]. As such, men withheart failure experienced an overnight increase in neck circumference seven timesgreater than women [11]. Recently, Yadollahi et al. studied the dynamics of fluid shift


Table 4. Mean and standard deviation of the variables within each cluster andassociated p-values for statistical differences between the clusters.

Cluster 1 Cluster 2 P-value

NFVst5 249.7±73.4 198.1±26.4 0.017LFVst5 2341±276 2050±225 0.005NCsp1 39.7±2.6 33.5±1.3 <0.001ΔNC90 0.64±0.40 0.40±0.28 0.136ΔNFV90 17.0±5.2 14.4±3.1 0.308ΔUAXSA90 −0.43±0.67 −0.18±0.76 0.148

in men and women lying awake in the supine position for 90 minutes to investigate thedifferences in fluid shift between the two sexes [34]. They found that men and womenexperienced a similar volume of fluid shift out of their legs, with men accumulatingmore fluid in their thorax and neck [34]. In addition, there was also a somewhatsignificant tendency for women to accumulate more fluid in their abdomen, comparedto men [34]. It has been suggested that since women have larger gonadal veins and thelarge venous plexus around the uterus [39,40], the venous pooling in the pelvic regionof women is greater than that of men [41]. Therefore, in women, the fluid shifting outof the legs may be accumulating more in the pelvis and reducing fluid redistributioninto the neck [11,34].

Our results comply with previous studies showing that sex has a significant effect onthe change in NC after 90 minutes. Clustering based on the subset of variables selectedfrom the PCA separated the data mainly based on sex. Even when sex was excludedfrom the clustering analysis, the data was grouped based on baseline NC and baselineLFV and NFV (significant contributors to the first PCA), which were significantlydifferent between the sexes (Table 1). In Figure 3, cluster 1 represents female subjectswith lower baseline NC, NFV and LFV, whereas cluster 2 represents male subjects withhigher baseline NC, NFV, and LFV. However, the stepwise regression model shows thatin addition to sex, LFVst5, NFVsp1, and NCsp1 also entered the model. Together, thesefactors accounted for 40% of variation in ΔNC90, as indicated by the partial R2 due tothese factors (Table 3). The inclusion of these variables, specifically baseline LFV,demonstrates that baseline LFV has an independent effect on ΔNC90, beyond the effectsof sex. This complies with previous studies showing that interventions reducingbaseline LFV, such as compression stockings [42–44], exercise [45], and diuretics [46],have led to less overnight increases in NC.

The change in NC during the supine period was also correlated with the change inNFV, demonstrating that the change in NC is indeed a reflection of the change in fluidaccumulating in the neck. This finding has been established by the previous work ofYadollahi et al. [34]. The correlation between ΔNFV90 and ΔUAXSA90 was notsignificant, but there was a trend to suggest that more fluid entering the neck during thesupine period corresponded to further narrowing of the UA-XSA (p = 0.087). Onelikely explanation for the lack of correlation between these variables is the shortduration (90 minutes), during which participants are laying supine. This did not allowenough time for the airway to narrow as a result of neck fluid accumulation.

Variables that were found to be predictive of both ΔNFV90 and ΔUAXSA90 werebaseline UAXSA and baseline NFV (standing and supine). These variables formedpositive relationships with ΔNFV90 and negative relationships with ΔUAXSA90.Therefore, increased baseline NFV and UAXSA were related to an increased neck fluidvolume and further narrowing of the upper-airway over the 90-minute supine period,both of which signal the accumulation of fluid in the neck. A likely explanation is thata higher baseline NFV and a narrower baseline UAXSA could be symptoms of havinga higher overall fluid retention at baseline. Thus, when lying supine, those with a greaterbaseline fluid retention experience an increased fluid shift, resulting in greater increasesin NFV and a narrowing of UAXSA over the 90-minute period.


The results of this study identify characteristics that could possibly foresee those atrisk of OSA through the fluid shift mechanism. Past studies have shown that anincreased overnight fluid shift from the legs and an increased overnight change in NCare both strongly related to an increased OSA severity [2]. This study identifiesphenotypic characteristics, specifically of the male sex, and increased baseline LFV, allof which are related to both an increased fluid shift from the legs and an increasedchange in NC when lying supine for a prolonged period, therefore possibly predictingthe likelihood of OSA through the fluid shift mechanism. Future work is aimed atexploring the capability of these characteristics and predicting those at risk of OSAthrough the fluid shift mechanism.

This study is subject to limitation, mainly imposed by the sensitivity of bioelectricalimpedance measurements to movement. Bioelectrical impedance measurements aresensitive to movement and body posture, and because our participants could not remainstill for more than 90 minutes, the duration of study was limited to just that. As a result,the fluid shift might be scaled down compared to overnight fluid shift. Our study wasperformed during wakefulness to limit involuntary body movements during sleep. Sincethe pattern of fluid redistribution out of the legs and into the neck could vary dependingon whether or not the subject was awake or sleeping, future work should include studiesspecifically during sleep. Finally, we did not examine correlations between baselinevariables and outcome variables in men and women separately due to the small numberof subjects in each group. In the future we aim to confirm our findings in larger groupsof men and women to gain a better understanding of fluid shift between the sexes. Inaddition, future work should adapt the models developed in our study using data fromovernight polysomnography – the gold standard for diagnosing sleep apnea. Modelsdeveloped with these data can be used to predict outcomes related to respiratorydisorders affected by fluid, such as OSA.

5. CONCLUSIONSIn conclusion, this study demonstrates that a baseline measure LFV is predictive of theamount of fluid leaving the legs and the change in NC after lying supine for 90 minutes.Our findings also suggest that with comparable changes in LFV over the supine period,men experience greater fluid accumulation in the neck, compared to women.Applications of these findings are in identifying characteristics of individuals at risk ofOSA through an overnight rostral fluid shift. There are also applications in thedevelopment of novel therapies aimed at reducing baseline LFV to minimize theassociated rostral fluid shift and accumulation in the neck during sleep. Such therapiesmay include reducing sedentary behaviors such as prolonged sitting [14] and wearingcompression stockings [42,43]. Future work should explore other interventions such asphysical activity [45, 47, 48] or electrical stimulation of the calf muscle pump [49] toreduce baseline leg edema and neck fluid accumulation in a population at risk for fluidshift related sleep apnea.

CONFLICT OF INTERESTThe authors indicated no potential conflicts of interest.


ABBREVIATIONSBMI Body mass indexNCsp1 Neck circumference at 1 minute supineUAXSAsp1 Upper airway cross-sectional area at 1 minute supineNFVst5 Neck fluid volume at 5 minutes standingLFVst5 Leg fluid volume at 5 minutes standingΔNFVP Change in neck fluid volume due to posture changeΔLFVP Change in leg fluid volume due to posture changeNFVsp1 Neck fluid volume at 1 minute supineLFVsp1 Leg fluid volume at 1 minute supineΔLFV90 Change in leg fluid volume after 90 minutes supineΔNC90 Change in neck circumference after 90 minutes supineΔNFV90 Change in neck fluid volume after 90 minutes supineΔUAXSA90 Change in upper airway cross-sectional area after 90 minutes supineLUA Length of the upper airway

REFERENCES[1] K. L. Chiu, C. M. Ryan, S. Shiota, P. Ruttanaumpawan, M. Arzt, J. S. Haight, C. T. Chan, J. S. Floras,

and T. D. Bradley. Fluid shift by lower body positive pressure increases pharyngeal resistance inhealthy subjects. Am. J. Respir. Crit. Care Med., 2006, 174(12):1378–1383.

[2] S. Redolfi, D. Yumino, P. Ruttanaumpawan, B. Yau, M. C. Su, J. Lam, and T. D. Bradley. Relationshipbetween overnight rostral fluid shift and Obstructive Sleep Apnea in nonobese men. Am. J. Respir.Crit. Care Med., 2009, 179(3): 241–246.

[3] S. Shiota, C. M. Ryan, K. L. Chiu, P. Ruttanaumpawan, J. Haight, M. Arzt, J. S. Floras, C. Chan, andT. D. Bradley. Alterations in upper airway cross-sectional area in response to lower body positivepressure in healthy subjects. Thorax, 2007, 62(10):868–872.

[4] M. C. Su, K. L. Chiu, P. Ruttanaumpawan, S. Shiota, D. Yumino, S. Redolfi, J. S. Haight, and T. D.Bradley. Lower body positive pressure increases upper airway collapsibility in healthy subjects. RespirPhysiol Neurobiol, 2008, 161(3):306–312.

[5] O. Friedman, T. D. Bradley, and A. G. Logan. Influence of lower body positive pressure on upperairway cross-sectional area in drug-resistant hypertension. Hypertension, 2013, 61(1):240–245.

[6] J. W. Shepard, D. A. Pevernagie, A. W. Stanson, B. K. Daniels, and P. F. Sheedy. Effects of changesin central venous pressure on upper airway size in patients with obstructive sleep apnea. Am. J. Respir.Crit. Care Med., 1996, 153(1):250–254.

[7] T. Kasai, S. S. Motwani, D. Yumino, J. M. Gabriel, L. T. Montemurro, V. Amirthalingam, J. S. Floras,and T. D. Bradley. Contrasting effects of lower body positive pressure on upper airways resistance andpartial pressure of carbon dioxide in men with heart failure and obstructive or central sleep apnea. J.Am. Coll. Cardiol., 2013, 61(11):1157–1166.

[8] M. C. Su, K. L. Chiu, P. Ruttanaumpawan, S. Shiota, D. Yumino, S. Redolfi, J. S. Haight, B. Yau, J.Lam, and T. D. Bradley. Difference in upper airway collapsibility during wakefulness between menand women in response to lower-body positive pressure. Clin. Sci., 2009, 116(9):713–720.

[9] L. H. White and T. D. Bradley. Role of nocturnal rostral fluid shift in the pathogenesis of obstructiveand central sleep apnoea. J. Physiol. (Lond.), 2013, 591(5):1179–1193.

[10] O. Friedman, T. D. Bradley, C. T. Chan, R. Parkes, and A. G. Logan. Relationship between overnightrostral fluid shift and obstructive sleep apnea in drug-resistant hypertension. Hypertension, 2010,56(6):1077–1082.


[11] T. Kasai, S. S. Motwani, D. Yumino, S. Mak, G. E. Newton, and T. D. Bradley. Differing relationshipof nocturnal fluid shifts to sleep apnea in men and women with heart failure. Circ Heart Fail, 2012,5(4):467–474.

[12] E. H. Starling. On the Absorption of Fluids from the Connective Tissue Spaces. J. Physiol. (Lond.),1986, 19(4):312–326.

[13] A. Krogh, E. M. Landis, and A. H. Turner. The movement of fluid throguh the human capillary wallin relation to venous pressure and to the collogid osmotic pressure of the blood. J. Clin. Invest., 1932,11(1):63–95.

[14] M. Pottier, A. Dubreuil, and H. Monod. The effects of sitting posture on the volume of the foot.Ergonomics, 1969, 12(5):753–758.

[15] R. L. Waterfield. The effect of posture on the volume of the leg. J Physiol, 1931, 72(1):121–131.

[16] J. B. Youmans, H. S. Wells, D. Donley, D. G. Miller, and H. Frank. The effect of posture (standing) onthe serum protein concentration and colloid osmotic pressure of blood from the foot in relation to theformation of edema. J. Clin. Invest., 1934, 13(3):447–459.

[17] P. Avasthey and E. H. Wood. Intrathoracic and venous pressure relationships during responses tochanges in body position. J Appl Physiol, 1974, 37(2):166–175.

[18] G. Baccelli, P. Pacenti, S. Terrani, M. Checchini, G. Riglietti, F. Prestipino, E. Omboni, F. Sardella, M.Catalano, and Z. Malacarne. Scintigraphic recording of blood volume shifts. J. Nucl. Med., 1995,36(11): 2022–2031.

[19] W. Hildebrandt, H. C. Gunga, J. Herrmann, L. Rocker, K. Kirsch, and J. Stegemann, Enhanced slowcaudad fluid shifts in orthostatic intolerance after 24-h bed-rest, Eur J Appl Physiol Occup Physiol,1994, 69(1):61–70.

[20] N. Terada and T. Takeuchi, Postural changes in venous pressure gradients in anesthetized monkeys.Am. J. Physiol, 1993, 264(1):H21–25.

[21] D. Yumino, S. Redolfi, P. Ruttanaumpawan, M. C. Su, S. Smith, G. E. Newton, S. Mak, and T. D.Bradley. Nocturnal rostral fluid shift: a unifying concept for the pathogenesis of obstructive andcentral sleep apnea in men with heart failure. Circulation, 2010, 121(14):1598–1605.

[22] B. E. Schroth. Evaluation and management of peripheral edema. JAAPA, 2005, 18(11):29–34.

[23] A. Yadollahi, F. Rudzicz, S. Mahallati, M. Coimbra, and T. D. Bradley. Acoustic Estimation of NeckFluid Volume. Ann Biomed Eng, 2014, 42(10):2132-2142.

[24] J. J. Fredberg, M. E. Wohl, G. M. Glass, and H. L. Dorkin. Airway area by acoustic reflectionsmeasured at the mouth. J Appl Physiol Respir Environ Exerc Physiol, 1980, 48(5):749–758.

[25] S. Demura, S. Sato, and T. Kitabayashi. Percentage of total body fat as estimated by three automaticbioelectrical impedance analyzers. J Physiol Anthropol Appl Human Sci, 2004, 23(3):93–99.

[26] F. Zhu, D. Schneditz, E. Wang, and N. W. Levin. Dynamics of segmental extracellular volumes duringchanges in body position by bioimpedance analysis. J. Appl. Physiol., 1998, 85(2):497–504.

[27] M. Y. Jaffrin and H. Morel. Body fluid volumes measurements by impedance: A review ofbioimpedance spectroscopy (BIS) and bioimpedance analysis (BIA) methods. Med Eng Phys, 2008,30(10):1257–1269.

[28] U. G. Kyle, I. Bosaeus, A. D. De Lorenzo, P. Deurenberg, M. Elia, J. M. Gomez, B. L. Heitmann, L.Kent-Smith, J. C. Melchior, M. Pirlich, H. Scharfetter, A. M. Schols, and C. Pichard. Bioelectricalimpedance analysis–part I: review of principles and methods. Clin Nutr, 2004, 23(5):1226–1243.

[29] A. De Lorenzo, A. Andreoli, J. Matthie, and P. Withers. Predicting body cell mass with bioimpedanceby using theoretical methods: a technological review. J. Appl. Physiol., 1997 82(5):1542–1558.

[30] D. Bracco, D. Thiebaud, R. L. Chiolero, M. Landry, P. Burckhardt, and Y. Schutz. Segmental bodycomposition assessed by bioelectrical impedance analysis and DEXA in humans. J. Appl. Physiol.,1996, 81(6):2580–2587.

[31] C. H. Ling, A. J. de Craen, P. E. Slagboom, D. A. Gunn, M. P. Stokkel, R. G. Westendorp, and A. B.Maier. Accuracy of direct segmental multi-frequency bioimpedance analysis in the assessment of total


body and segmental body composition in middle-aged adult population. Clin Nutr, 2011,30(5):610–615.

[32] A. Tagliabue, A. Andreoli, M. Comelli, S. Bertoli, G. Testolin, G. Oriani, and A. De Lorenzo.Prediction of lean body mass from multifrequency segmental impedance: influence of adiposity. ActaDiabetol, 2001, 38(2):93–97.

[33] F. Zhu, M. K. Kuhlmann, P. Kotanko, E. Seibert, E. F. Leonard, and N. W. Levin. A method for theestimation of hydration state during hemodialysis using a calf bioimpedance technique. Physiol Meas,2008, 29(6):S503–516.

[34] A. Yadollahi, B. Singh, and T. D. Bradley. Investigating the Dynamics of Supine Fluid RedistributionWithin Multiple Body Segments Between Men and Women. Ann Biomed Eng, 2015, 43(9):2131–2142

[35] M. Fenech and M. Y. Jaffrin. Extracellular and intracellular volume variations during postural changemeasured by segmental and wrist-ankle bioimpedance spectroscopy. IEEE Trans Biomed Eng, 2004,51(1):166–175.

[36] XITRON, HYDRA ECF/ICF (Model 4200), Bio-impedance spectrum analyzer, For measuringintracellular and extracellular fluid volumes. Operating manual. XITRON Technologies Inc., 2007.

[37] I. Jolliffe, Principal component analysis. Wiley Online Library, 2002.

[38] J. A. Hartigan and M. A. Wong. A k-means clustering algorithm: Algorithm AS 136. Applied statistics,1979, 28(1):100–108.

[39] N. E. Ahlberg, O. Bartley, and N. Chidekel. Right and left gonadal veins. An anatomical and statisticalstudy. Acta Radiol Diagn (Stockh), 1966, 4(6):593–601.

[40] H. Gray, The veins of the abdomen and pelvis, in Anatomy of the Human Body, 29th ed., C. Goss, Ed.Philadelphia, PA: Lea & Febiger, 1973, 709–715.

[41] D. D. White and L. D. Montgomery. Pelvic blood pooling of men and women during lower bodynegative pressure. Aviat Space Environ Med, 1996, 67(6):555–559.

[42] S. Redolfi, I. Arnulf, M. Pottier, J. Lajou, I. Koskas, T. D. Bradley, and T. Similowski. Attenuation ofobstructive sleep apnea by compression stockings in subjects with venous insufficiency. Am. J. Respir.Crit. Care Med., 184(9):1062–1066.

[43] S. Redolfi, I. Arnulf, M. Pottier, T. D. Bradley, and T. Similowski. Effects of venous compression ofthe legs on overnight rostral fluid shift and obstructive sleep apnea. Respir Physiol Neurobiol, 2011,175(3):390–393.

[44] L. H. White, O. D. Lyons, A. Yadollahi, C. M. Ryan, and T. D. Bradley, Effect of below-the-kneecompression stockings on severity of obstructive sleep apnea, Sleep Med., 2015, 16(2):258–264.

[45] S. Redolfi, M. Bettinzoli, N. Venturoli, M. Ravanelli, L. Pedroni, L. Taranto-Montemurro, I. Arnulf,T. Similowski, and C. Tantucci. Attenuation of obstructive sleep apnea and overnight rostral fluid shiftby physical activity. Am. J. Respir. Crit. Care Med., 2015, 191(7):856–858.

[46] T. Kasai, T. D. Bradley, O. Friedman, and A. G. Logan. Effect of intensified diuretic therapy onovernight rostral fluid shift and obstructive sleep apnoea in patients with uncontrolled hypertension. J. Hypertens., 2014, 32(3):673–680.

[47] I. H. Iftikhar, C. E. Kline, and S. D. Youngstedt. Effects of exercise training on sleep apnea: a meta-analysis. Lung, 2014, 192(1):175–184.

[48] C. E. Kline, E. P. Crowley, G. B. Ewing, J. B. Burch, S. N. Blair, J. L. Durstine, J. M. Davis, and S.D. Youngstedt. The effect of exercise training on obstructive sleep apnea and sleep quality: arandomized controlled trial. Sleep, 2011, 34(12):1631–1640.

[49] A. A. Goddard, C. S. Pierce, and K. J. McLeod. Reversal of lower limb edema by calf muscle pumpstimulation. J Cardiopulm Rehabil Prev, 2008, 28(3):174–179.


International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation http://www.hindawi.com

Journal ofEngineeringVolume 2014

Submit your manuscripts athttp://www.hindawi.com

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation http://www.hindawi.com

Volume 2014

The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014

SensorsJournal of


Modelling & Simulation in EngineeringHindawi Publishing Corporation http://www.hindawi.com Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks


Documents

Predicting Neck Fluid Accumulation While Supine · leg fluid volume (LFV) between 100 and 300 ml [14–16]. When lying down, the reverse occurs; hydrostatic pressure in the capillaries